Effective model for pneumonia detection from chest x-rays using deep convolutional neural networks/
Pneumonia is a disease which occurs in the lungs due to bacterial, viral or fungal infection and causes lung alveoli to fill with pus or fluid. Chest x-ray is the most common diagnostic tool for pneumonia. However, because of several other conditions in the lungs such as volume loss, bleeding, fluid...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Kuala Lumpur :
Kulliyyah of Engineering,International Islamic University Malaysia,
2020
|
| Subjects: | |
| Online Access: | http://studentrepo.iium.edu.my/handle/123456789/10129 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Pneumonia is a disease which occurs in the lungs due to bacterial, viral or fungal infection and causes lung alveoli to fill with pus or fluid. Chest x-ray is the most common diagnostic tool for pneumonia. However, because of several other conditions in the lungs such as volume loss, bleeding, fluid overload, lung cancer or post-radiation or surgical changes, the diagnosis of pneumonia in chest x-rays becomes very complicated. Therefore, there is urgent need for computer aided diagnosis systems to assist clinicians in making better decisions. In this work, a deep convolutional neural network, ResNet-50 architecture, is proposed and is trained using transfer learning technique. A pre-trained model on ImageNet dataset is used and with the use of transfer learning, ResNet-50 model is trained for binary classification of chest x-ray images into pneumonia and non-pneumonia. Two datasets have been used and the ResNet-50 model was implemented on both the datasets. The model achieved an accuracy of 96.76% with RSNA dataset and 94.06% with Chest X-ray Image (CXI) dataset. RSNA dataset despite having almost five times more images than CXI dataset took very less time for training. Also because of the use of transfer learning technique both the ResNet-50 models were able to learn the significant features of pneumonia with only 50 % training. The proposed ResNet-50 model gave an accuracy of 96.76 %, however, the model can be improved by using more deeper networks. Furthermore, this work could be extended to detect and classify x-ray images consisting of both lung cancer and pneumonia. |
|---|---|
| Physical Description: | xiv, 112 leaves : colour illustrations ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 95-105). |